The calculations to estimate incidence, hospitalizations, and deaths are described below and in the Figure. Details of the data sources and the simulation technique used to account for uncertainty are shown in the Appendix.

We considered the definition of foodborne to include any infectious gastroenteritis caused by eating food, including food contaminated just before eating. The proportion of infectious gastroenteritis cases that are due to foodborne transmission in the community and their proportion among hospitalizations and deaths, were assessed by using data from multiple Australian sources about individual known pathogens.

Gastroenteritis caused by known pathogens was studied to estimate the proportion of all infectious gastroenteritis that is foodborne. Of 25 pathogens with the potential for foodborne transmission, 16 (those listed in Table 1) were considered relevant. Pathogens not considered relevant were either not foodborne (Clostridium difficile), did not cause gastroenteritis (Brucella spp., Listeria spp., Toxoplasma gondii, hepatitis A virus), or were only acquired overseas (Salmonella Typhi, Vibrio cholerae, Cyclospora cayetanensis, Trichinella spiralis).

For each of the 16 pathogens, yearly estimates were made of the total number of cases of gastroenteritis in the community; this number was based on data collected in the National Notifiable Diseases Surveillance System, published results from the Water Quality Study conducted in Melbourne in 1998 (17,18), laboratory data, or outbreak data. Necessary adjustments were made for underreporting in the Australian surveillance system, incomplete population coverage, and the proportion of infections acquired overseas (Appendix). We also estimated, for each of the pathogens, the total number of hospitalizations and deaths based on data from the National Hospital Morbidity Database (NHMD) as well as the proportion of cases due to foodborne transmission, which was estimated from outbreak data, the literature, and because Australian data were lacking, a Delphi process involving 10 foodborne disease experts in Australia. The number of foodborne episodes for each pathogen was then obtained by multiplying the estimate of the foodborne proportion by the estimate of the total number of cases, hospitalizations, and deaths. The overall proportion of infectious gastroenteritis in the community due to foodborne transmission was then estimated by dividing the sum of the foodborne cases due to the 16 known pathogens by the sum of all cases due to the 16 known pathogens (Figure). The equivalent calculation was also done for hospital admissions due to known pathogens to give the proportion of hospitalizations for infectious gastroenteritis due to foodborne transmission.

We assumed that the proportion of gastroenteritis that is foodborne is the same among cases caused by known pathogens as among those caused by unknown pathogens. Adjustments were made for the proportion estimated as acquired overseas for certain pathogens (Appendix).

The estimates from this study demonstrate the considerable prevalence of foodborne disease in contemporary Australia and justify the attention given to foodborne disease surveillance and food safety. The uncertainty estimates indicate that even the lower boundary of the credible interval is still high, with at least 4 million cases of foodborne gastroenteritis, and possibly as many as 7 million per year. This means that on average, every Australian can expect to experience an episode of foodborne illness about every 3 to 4 years. Hospitalizations are uncommon at 8 per 10,000 people each year, and ≈4 deaths per million persons occur per year.

Similar studies have been done in United States (6) and the United Kingdom (7). The Australian estimate of incidence is remarkably similar to that reported for the United States, but higher than in the United Kingdom. In the United States, 36% of all gastroenteritis was estimated to be due to foodborne transmission, and incidence was estimated at 0.28 cases per person per year. In the United Kingdom, 26% of gastroenteritis was estimated to be due to foodborne transmission, and incidence was estimated at 0.04 cases per person per year in 1995. The importance of using a standardized method when comparing results of the amount of foodborne gastroenteritis across countries or times cannot be overemphasized. Evidence suggests that a prospective cohort study design may produce a lower incidence of community gastroenteritis than a cross-sectional design. The UK study included a quality control substudy to compare the incidence based on a retrospective recall method with incidence from a prospective diary method; the estimates of incidence were 0.6 and 0.2 cases per person per year, respectively (14). Prospective studies that require participants to supply a stool sample every time they report gastroenteritis might tend to cause an underestimate because of unwillingness to provide a sample; on the other hand, in a retrospective recall method, respondents might "telescope" events into a shorter time frame. A prospective study done in the Netherlands (22) also found a lower incidence than that seen in the United Kingdom. Other variations in methods also exist across countries, such as differences in surveillance systems and the quality of outbreak data available to estimate the proportion of cases that are foodborne. Not only may the study design influence the final estimate, but also the definition of gastroenteritis. Even a seemingly small difference in the definition of gastroenteritis can lead to a considerable difference in the final estimates (23).

The definition of community gastroenteritis used in this Australian study refers to moderate-to-severe illness, with at ≥3 loose stools or ≥2 episodes of vomiting in a single day. To improve the specificity of our definition for enteric illness, we excluded patients with concomitant respiratory symptoms unless they had more severe symptoms of diarrhea or vomiting. Previous studies have found similarly high rates of respiratory symptoms amongst cases of gastroenteritis (24). A definition inclusive of milder illness would lead to a higher estimate of foodborne gastroenteritis, and a definition that included only more severe illness would lead to a lower estimate.

We also took account of those with concurrent respiratory symptoms in our definition of community gastroenteritis, although most studies estimating the amount of gastroenteritis have not considered this. The United States study (6) adjusted for those with respiratory illness by excluding a proportion of case-patients who were thought likely to have symptoms secondary to respiratory infections rather than a primary enteric infection. The UK definition of gastroenteritis was different from the Australian definition in several ways. While differing arguments can be raised about the best definition of gastroenteritis, the main concern is to have a consistent, reasonable definition for comparative purposes.

Of the 5.4 million (95% CrI 4.0–6.9 million) cases of foodborne gastroenteritis, 28% were attributed to known pathogens. This finding compares with 18% in the United States (6) and 41% in the United Kingdom (7). The Australian data used to estimate the pathogen-specific numbers of community cases of gastroenteritis were variable in quality. Salmonella notifications have been relatively stable over the last 5 years, and characteristics of this illness are fairly well understood. In comparison, reports of illness due to Campylobacter have increased steadily during the same time (25). This finding could be due to reporting artifacts or increasing infection rates in the community. The diagnostic laboratory tests have not changed appreciably during this time.

The pathogen-specific estimates in this study that most influenced the final estimate of the proportion of gastroenteritis that is foodborne were those for norovirus and enteropathogenic E. coli, as these accounted for the largest numbers. The estimates for both were determined from a high-quality longitudinal study (17). Nevertheless, the sample was limited to a specific subpopulation and geographic location. The high proportion of E. coli is similar to findings in the United Kingdom, although we estimated that 50% of cases caused by this pathogen were foodborne compared to 8% of cases in the UK assessment (7) and 30% in the United States (6). In recent years, the capacity of laboratories to identify noroviruses with polymerase chain reaction tests has improved considerably, and this virus is likely to become increasingly recognized (26).

Factors were used to adjust for underreporting when using data from outbreaks and surveillance. Further studies are needed to give more robust estimates of the level of underreporting compared with the true level in the community. The estimates of the proportion of illness due to foodborne transmission for specific pathogens relied largely on outbreak data and opinions of foodborne diseases experts. Outbreak data can be very sensitive to the outcomes from larger events, which could bias the estimate of the proportion foodborne in either direction (27). Foodborne disease experts' experience was based on pathogen characteristics in the laboratory, results of outbreak investigations, and knowledge from case-control studies of sporadic infections. For pathogens estimated to have a large number of cases, such as norovirus, the estimate of the proportion thought to be foodborne can influence the final estimate. Both the UK and Australian estimates were based on outbreak data, but only 11% of Norwalk-like virus (caliciviruses) gastroenteritis was ascribed to foodborne transmission in the UK study, compared with 40% in the United States, and 25% in Australia (6,7). These individual estimates had some influence on the final estimates of the proportion of all gastroenteritis that is foodborne.

Hospital data in Australia are fairly complete, and only a few hospitals, mostly private, have not contributed records of all admissions to the national database in the last decade (21). Coding of admissions varies over time and place, but a patient with gastroenteritis is likely to be coded for this condition in the first 10 diagnoses (28). Approximately two thirds of diagnoses were coded as the main reason for admission. Additional diagnoses may represent cases with complications or comorbidity that took precedence in the order of coding or cases acquired in the hospital. Some deaths due to gastroenteritis may have occurred in nursing homes, which were not included. Among the known pathogens, bacterial infections accounted for >90% of hospital admissions in Australia, which is similar to the proportion in the United Kingdom (7) but higher than the 60% estimated in the United States (6). Campylobacteriosis followed by salmonellosis accounted for most admissions due to bacterial infections in Australia and the United Kingdom; in the United States this order was reversed. These illnesses are important when considering the severe end of the spectrum of foodborne gastroenteritis.

We used the best available Australian data to conduct this study, but as experienced by others conducting research overseas, the quality of the data inevitably varied. Since the quality of the data cannot be easily improved, we chose to provide estimates that reflect the true state of uncertainty of the data by using a simulation technique that can be easily applied. Taking account of uncertainty informs the data users, including policy makers, that a very precise estimate is not possible. An appreciation of the degree of confidence that can be placed in an estimate is an important part of the responsible presentation of results that may have considerable effects at a policy level. With these stipulations, we are confident that the level of foodborne gastroenteritis is high in Australia. In the future, improvements in data completeness and quality would enhance the robustness of the calculations, but estimates of uncertainty are likely to remain an important component of the results.